Containers (open discussion)

CMAF + Low latency

A bit of CMAF history.

CMAF consists of ISOBMFF + Branding + media profiles + presentation profiles.

Akamai interested in low latency CMAF. Will explained how low latency (< 500ms may be attained).

RTP more mature, but CMAF + DASH provides a much fuller feature set.

CMAF only MPEG codecs, but working on adding Open codec profiles.

Matroska may already have solved a some of the low-latency problems which CMAF is trying to address.

YouTube? is planning on adopting CMAF, and currently supports WebM?/Matroska. CMAF fixes some of the timing issues of ISO-BMFF.

The ecosystem is converging on CMAF, and is based on open, royalty free standards (sans Codecs).

YouTube? does not feel there is much impetus in putting up a fight for MKV vs MPo?.

Youtube has found MKV to be harder to work with compared to MP4?.

Other codecs coming to MP4?: AV1?, Opus. Both also available for MKV encapsulation.

The new audio codecs are multi-track. They may require multiple streams which get reproduced together.

Not currently supported by MSE. There is ongoing discussion on how to support this.

Dolby Vision: Addinteg HDR layer on top of AVC, which was problematic, so they have switched to a single-layer approach.

Some web-based devices are supporting Dolby Vision, namely Edge. However, HDR dynamic mastering metadata is not supported anywhere yet.

Chrome supports HDR10? with VPno9?. However HDR monitors are in short supply, if available at all. HDR10? within an ISO container will probably come with AV1? for YouTube?.

Low-Latency Streaming

Low-latency modes do not work necessarily with b-frames due to latency introduced by the platform's video decoder holding on to frames. So sometimes the issue is more a codec issue than a container issue.

Firefox will drain the decoder when it runs out of data.

Providers are working towards sub-second latency, but it is unrealistic in some applications and platforms. Creators sometimes have a call/response type of application. For this type of use case, latency is very important.

How to solve the low-latency issue on 3G or similar type networks? TCP makes it hard to do synchronous delivery over 3G because the medium can be unpredictable. Sometimes the solution is to add infrastructure to drop packets if not delivered on time. Some presentations will provide streams which can be delivered low-latency, and others offer no such expectation over higher-latency channels. The player has to make the right choice.

Would IPV6? perhaps address some of these issues with QOE and reserved bandwidth? Not for quite a while, when all the operators cooperate to make it work.

There are many algorithms to determine the delivery policy, and more algorithms in the CDN and network to enforce them, but no standardization at this point.

Matroska Standardization